Which texture format uses least video memory?

[Mel]

I was about to recomend that. Texture atlases have a great advantage. A model with the textures set into an atlas have the advantage that the atlas is (can be) just one single big texture, and thus, the meshes that use it can be stored as a single meshbuffer, with the advantages this can also provide. Of course, the levels should be exported as a single model, and you have to map the material properties within the atlas, instead of per model, but the number of draw calls also reduce.

It is still a large texture, and it may use much ram, but you are using less texture resources when rendering, and this can improve performance too.

[Lonesome Ducky]

Yep, texture atlases are great for performance. But remember that they suffer greatly from mipmap bleeding, and solutions to this are not always trivial.

[roelor]

I don't know if my method will work very well, but I am using 2 scenemanagers for the rendering of my visuals (I haven't come very far tho) this in theory should let you display low resolution models and textures far away and let you show high detail stuff up close, the only thing you have to do is assign the low detail models etc to the low detail scene manager then load in not loaded stuff as you get close to them.

[devsh]

convert your textures to YUVA and split them into 2 separate planes (textures) one R8G8/LUM_ALPHA (2 channel) texture to code luminosity and alpha at full res and another for cR and cB. The good thing here is you usually do a 4:2:2 ratio where the YA texture is full res (1024x1024 for example) and the UV texture is quarter res (512x512) YOU GET DECENT LOOKING NON-PIXELATED IMAGES LIKE THAT!!! Bink and H264 do it!!! You just need a shader to decode the planes into RGBA... also this is the format best suited for texture streaming because you should care about loading the lum_a first and then upscale the UV

[3DModelerMan]

I'm actually writing a patch for DXT1 and DXT3 compressed textures in Irrlicht right now.

[ACE247]

NICE! This is an old topic, but nonetheless, go on.

[Cube_]

old?
october 18 isn't that long ago...

[serengeor]

old?
october 18 isn't that long ago...

Nope, it's 'Sat Oct 16, 2010 9:26 pm'

[hendu]

convert your textures to YUVA and split them into 2 separate planes (textures) one R8G8/LUM_ALPHA (2 channel) texture to code luminosity and alpha at full res and another for cR and cB. The good thing here is you usually do a 4:2:2 ratio where the YA texture is full res (1024x1024 for example) and the UV texture is quarter res (512x512) YOU GET DECENT LOOKING NON-PIXELATED IMAGES LIKE THAT!!! Bink and H264 do it!!! You just need a shader to decode the planes into RGBA... also this is the format best suited for texture streaming because you should care about loading the lum_a first and then upscale the UV

If you have fine color details, especially red, the 4:2:2 or 4:2:0 formats degrade quality visibly.

[Virion]

you can try convert your texture to DDS after compression. DDS is slightly faster than normal formats because it can be read directly by the GPU.

[Cube_]

old?
october 18 isn't that long ago...

Nope, it's 'Sat Oct 16, 2010 9:26 pm'

nope, it is

Mon Oct 18, 2010 6:02 pm

**time zone differences you know **